0
$\begingroup$

As you know increase a data samples on training is a way to preventing of over fitting.

I'm working on an UCI data set with 198 samples and 34 feature this is my data set's dimension and I wanted to know what's the meaning of increasing our data to prevent of over fitting, I mean I should make a change on selecting test and try data size or what?

$\endgroup$
2
$\begingroup$

Since you are working with a given data set, you can't increase the amount of data, so you will have to use simpler models.

How much simpler? That depends on the model. There are some rules of thumb, e.g., for linear regression you want at least 10 observations for each independent variable; for logistic regression you want at least 10 observations in the smallest class for each independent variable and so on, but they are only rules of thumb (and there are others that recommend different limits).

This isn't solvable by changing the train and test sizes as increasing the size of one will decrease the size of the other.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Will leave-one-out cross-validation indicate the degree of "overfitting"? $\endgroup$ – German Demidov Nov 25 '19 at 14:18
1
$\begingroup$

It means literally going and getting more data. As in I took a survey of 100 people, and then decided I needed more data so I surveyed 50 more. If you’re using UCI data, increasing your sample is not possible.

| cite | improve this answer | |
$\endgroup$
1
$\begingroup$

If it's classification problem you can try to resample the data. Here's a lib with most of the popular methods.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.